Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76591.63094128086,
-      "required_pages_per_second_upper_estimate": 0.005298786745586228
+      "required_token_positions_per_second": 77242.9775793777,
+      "required_pages_per_second_upper_estimate": 0.005343848417342188
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9573.953867660108,
-      "required_pages_per_second_upper_estimate": 0.004008414557498086
+      "required_token_positions_per_second": 9655.372197422212,
+      "required_pages_per_second_upper_estimate": 0.004042502711961416
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19147.907735320216,
-      "required_pages_per_second_upper_estimate": 0.003988004216982843
+      "required_token_positions_per_second": 19310.744394844423,
+      "required_pages_per_second_upper_estimate": 0.004021918798870245
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28721.861602980323,
-      "required_pages_per_second_upper_estimate": 0.0017218317473553326
+      "required_token_positions_per_second": 28966.116592266633,
+      "required_pages_per_second_upper_estimate": 0.001736474461007273
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28721.861602980323,
-      "required_pages_per_second_upper_estimate": 0.0022329984055139297
+      "required_token_positions_per_second": 28966.116592266633,
+      "required_pages_per_second_upper_estimate": 0.0022519881565667846
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28721.861602980323,
-      "required_pages_per_second_upper_estimate": 0.3684132015107595
+      "required_token_positions_per_second": 28966.116592266633,
+      "required_pages_per_second_upper_estimate": 0.3715462422527499
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1049930411639366,
-    "producer_p95_seconds": 0.27777168399916263,
-    "gpu_update_seconds": 0.9626116991362115,
-    "producer_batches_per_second": 9.524440752588504,
-    "gpu_batches_per_second": 1.0388404804318694,
-    "producer_headroom": 9.168338096171397,
+    "producer_batch_seconds": 0.19855239045315898,
+    "producer_p95_seconds": 0.2911086139974941,
+    "gpu_update_seconds": 0.9544945354318382,
+    "producer_batches_per_second": 5.036454095151842,
+    "gpu_batches_per_second": 1.0476749346161254,
+    "producer_headroom": 4.807267911775736,
     "sustainable": true,
-    "buffer_seconds": 60.64453704558132,
+    "buffer_seconds": 60.1331557322058,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323065.54371093906,
+    "generation_bytes_per_second": 362914.0698987244,
     "upload_bytes_per_second": 4926004.937585538,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.32234509941225,
+    "estimated_full_shards_per_hour": 77.8729111930971,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 463200, "loss": 0.5854924321174622, "lr": 0.0003, "grad_norm": 0.14913088991432086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67798604936, "tokens_per_second": 152690.07047614496, "data_wait_s": 1.2446002074284479e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 463300, "loss": 0.5862105488777161, "lr": 0.0003, "grad_norm": 0.15822271236661156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67813291840, "tokens_per_second": 152541.32047891046, "data_wait_s": 8.896997314877808e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 463400, "loss": 0.6133775115013123, "lr": 0.0003, "grad_norm": 0.1669339987417047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67828016244, "tokens_per_second": 153073.8088861254, "data_wait_s": 1.0432002454763278e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 463500, "loss": 0.5764064788818359, "lr": 0.0003, "grad_norm": 0.15915337962806428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67842717702, "tokens_per_second": 152862.0500351235, "data_wait_s": 1.4067001757211983e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b7aa835a2e6b919b766e4da08d138b7cbf6917bad6cd5094df910e77a35e3d3
-size 494673
+oid sha256:eb99357abc3eccd6bc565fd24ceab156d7e772b96ee776ac0546fa452b08406f
+size 496791